How Stratum works.

The technical architecture behind the calibration. Written for the deeply curious.

01

Adaptive Load methodology.

Traditional training metrics fall into a few buckets. Tonnage counts total weight moved. INOL accounts for intensity relative to maxes. RPE-based systems treat reported effort as the primary measurement. Heart rate zones quantify conditioning load. Each captures something real. Each misses something too.

Tonnage doesn’t distinguish between systemic and local fatigue. A leg press at high tonnage produces less central fatigue than a back squat at the same tonnage. INOL accounts for intensity but doesn’t factor in positional demand. RPE relies on accurate self-report and varies wildly across lifters with the same actual fatigue state. Heart rate metrics work for conditioning but don’t translate to strength work.

Adaptive Load is Stratum’s attempt to combine the signals into a unified per-session fatigue measurement for strength work. The calculation:

Position multiplier accounts for whether the movement is a primary competition lift, a closely-related variant, or an accessory pattern. Axial multiplier accounts for spinal loading. Stance multiplier accounts for unilateral vs bilateral demand.

The result is a per-session fatigue value that can be summed for daily AL, rolled up to weekly AL, compared against historical averages, and tracked against personalized thresholds. The per-athlete piece matters most. AL thresholds are calibrated against the athlete’s own history. What’s a “heavy” AL session for a novice is different from what’s heavy for an intermediate. Stratum learns where each athlete’s heavy zone actually sits, then programs around that.

Conditioning work is captured in the same training plan via interval programming and ATL/CTL/TSB fitness modeling. The mechanical fatigue from strength work and the metabolic fatigue from conditioning are combined into a single per-athlete readiness signal using sport-derived modality weights. A pure powerlifter sits closer to 1.0/0.0 strength-to-cardio. A multi-modal athlete training for a hybrid event sits closer to 0.7/0.3. An endurance athlete with strength accessory work might weight closer to 0.3/0.7. The relative importance of mechanical vs metabolic fatigue varies by athlete and training goal — a universal formula would either oversimplify for some or distort for others. Weights are derived automatically from the athlete’s sport type and training history. Manual per-athlete override is on the roadmap.

This isn’t a new equation. The components draw from established sports science — Brzycki for e1RM, axial loading research from the early 2000s, periodization frameworks from Verkhoshansky and Zatsiorsky, training stress methodology from the endurance world. The contribution is integration: combining the signals into a single per-athlete adaptive number that’s actionable across modalities.

02

Subjective Recovery Accuracy.

Athletes self-report readiness through pre-session check-ins: sleep, soreness, motivation, life stress. Most platforms take those scores at face value and feed them into recommendations.

Stratum tracks a second metric: how accurate those check-ins actually are.

After each session, the system compares predicted performance (derived from the check-in) against measured performance (derived from the logged session). The gap between prediction and reality, accumulated over time, becomes the athlete’s Subjective Recovery Accuracy score.

An athlete with high SRA is reporting recovery accurately. Their check-ins predict their performance well. Programming can trust their reports and adjust intensity based on them.

An athlete with lower SRA — often the case during high-stress life periods, or early in calibration — has a meaningful gap between subjective and objective state. Their check-in says they’re at 6/10 readiness, but they hit prescribed work without issue, session after session. The system learns this pattern, weights their objective performance more heavily in programming decisions, and surfaces the gap to the coach.

This works the same way across modalities. A hypertrophy athlete with low SRA gets volume adjusted based on actual lift performance rather than self-reported fatigue. A sport athlete with high SRA can be trusted when they report needing a lighter conditioning day. The methodology is modality-agnostic — it tracks how well any given athlete’s subjective state predicts their objective performance.

The calibration window is empirical. Below a minimum sample of paired check-in / performance datapoints, the correlation isn’t trustworthy and the system explicitly says so — surfacing a “calibrating” state in the coach interface rather than projecting false confidence. Above that threshold, SRA stabilizes enough to use as a programming input. Athletes who haven’t crossed the threshold yet are flagged as calibrating so coaches know to treat recommendations with appropriate uncertainty.

A daily biofeedback log extends this signal beyond session check-ins. Athletes log sleep hours, stress, mood, appetite, joint pain location and severity, HRV, and resting heart rate in under 60 seconds each morning. These readings feed the subjective accuracy learner as additional context between training sessions — independent of the pre-session check-in and captured without friction at the start of the training day.

03

Volume landmarks.

MEV (minimum effective volume), MAV (maximum adaptive volume), and MRV (maximum recoverable volume) are tracked per muscle, not per lift.

The distinction matters across all modalities. A high-volume squat day hits quads hard but barely touches hamstrings. A high-volume deadlift day hits hamstrings, glutes, and erectors but minimal quad work. Programming volume by lift conflates muscle group responses. Programming volume by muscle keeps each tracking accurately.

Modality-specific defaults exist for powerlifting (lower body emphasis, MRV ranges tuned for SBD-adjacent muscles), S&C (broader distribution accounting for sport demands), and hypertrophy (higher MRV ranges across the board for muscle growth response). Defaults can be overridden per-athlete based on training history and individual response.

The coach inbox flags volume status per muscle: under MEV (need to add work), MEV-MAV (productive range), MAV-MRV (peak fatigue zone, watch for overreaching), over MRV (overreaching, deload imminent). This is the volume version of the readiness check — a system-level warning that programming decisions need to factor in.

Overreaching detection runs as a separate 13-signal composite score across the block. Signals include session AL trajectory, completion rate decline, RPE drift, HRV deviation from personal baseline, and daily readiness trends across sleep, stress, and joint pain severity — weighted against the athlete’s historical block tolerance. The score is gated: below a minimum check-in threshold, the system surfaces an explicit “insufficient data” state rather than projecting a risk score from noise. Above that threshold, the composite flags early overreaching before performance drops are measurable.

04

Self-healing exercise library.

Coaches add custom exercises regularly. “Pin squat to 14-inch box.” “SSB front-loaded zercher.” “Bench with chains and bands above 70 percent.” Most platforms either reject custom names, fail to categorize them, or break analytics when they encounter unfamiliar movements.

Stratum’s exercise library auto-categorizes custom additions against a taxonomy of movement patterns, muscle groups, and known variations. A “pin squat” is categorized as a squat variation with the appropriate volume credit toward quad / glute / lower back landmarks. Custom names work, and analytics stay accurate.

When the system encounters something truly novel, it flags it for review rather than silently failing. Unrecognized exercises are automatically classified using the rule-based engine and flagged for internal review — no coach input required for standard movements. The library accumulates accurately as athletes train, and corrections inform future volume credit for that athlete and surface to us for taxonomy improvements over time.

05

What Stratum isn’t.

We don’t claim science we haven’t earned. The methodology above draws from established research, validated frameworks, and ten years of coaching practice. It doesn’t pretend to be more than that.

We don’t pretend AI generates programs. Humans build programs. The platform helps them build better ones, faster, with more data to inform the decisions. Coaches stay in the loop on every programming decision.

We don’t sell certainty about training. We help athletes and coaches make better decisions with the data they have. The platform is a tool for thinking, not a replacement for thinking.

We also don’t ship claims we haven’t built. If something on this page reads as concrete capability, it’s in the product right now. If you find a gap between what we describe and what you encounter using the platform, tell us — that’s a bug, not a feature.